Yujia Zhai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Yujia Zhai has a total of 46 co-authors. Here are the top co-authors by number of shared publications:
- Ying Ding
University of Texas at Austin, Austin, USA
6 shared publications Shaojing Sun
School of Journalism Fudan University Shanghai China
3 shared publications- Shaojing Sun
Fudan University
3 shared publications Ying Ding
School of Information University of Texas at Austin Texas USA
3 shared publications- Fang Wang
Nankai University Tianjin China
2 shared publications - Hezhao Zhang
Department of Computer Science, The University of Sheffield, Sheffield, UK
2 shared publications
Latest Publications
- Open Access
- 22 April 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
From LLMs to AI agents: a systematic benchmark for SAO structure extraction in patent analytics
The continuous growth of patent literature has placed new demands on technological innovation and automated processing, particularly in the task of extracting Subject-Action-Object (SAO) relationships. This study constructs an information extraction benchmark framework tailored for the collaborative robotics domain, comparing three large language models—GPT-4o, Gemini 1.5 Pro, and Claude 3.5 Sonnet—and their corresponding agents. The experimental design employs a closed-loop “reflection and refinement” model, integrating the thought chain and ReAct method strategies, and configures decoding parameters to prioritize factual quality over linguistic diversity in extraction results. The experimental results indicate that the agent architecture has a positive impact on the performance of all models. Among them, the Gemini 1.5 Pro agent performed the best, achieving an F1 score of 89.58% (95% confidence interval: [87.9%, 91.2%]) and a BERTScore of 87.73%. A t-test statistical significance analysis based on 501 independent test sentence samples confirmed that the agent significantly outperforms baseline large language models, particularly in the parsing of recursively nested clauses and the filtering of legal jargon. Resource consumption analysis indicates that although the agent framework results in a 40- to 60-fold increase in computational cost—consuming an average of 11,000 to 15,000 tokens per sentence—this overhead is justified in specific patent analysis scenarios due to its superior recall performance and risk mitigation effects. Furthermore, by classifying typical failure modes and proposing a hierarchical deployment strategy, this study provides a methodological framework and empirical foundation for deploying and applying the agent in specific technical domains.
- 22 April 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessFrom LLMs to AI agents: a systematic benchmark for SAO structure extraction in patent analytics
The continuous growth of patent literature has placed new demands on technological innovation and automated processing, particularly in the task of extracting Subject-Action-Object (SAO) relationships. This study constructs an information extraction benchmark framework tailored for the collaborative robotics domain, comparing three large language models—GPT-4o, Gemini 1.5 Pro, and Claude 3.5 Sonnet—and their corresponding agents. The experimental design employs a closed-loop “reflection and refinement” model, integrating the thought chain and ReAct method strategies, and configures decoding parameters to prioritize factual quality over linguistic diversity in extraction results. The experimental results indicate that the agent architecture has a positive impact on the performance of all models. Among them, the Gemini 1.5 Pro agent performed the best, achieving an F1 score of 89.58% (95% confidence interval: [87.9%, 91.2%]) and a BERTScore of 87.73%. A t-test statistical significance analysis based on 501 independent test sentence samples confirmed that the agent significantly outperforms baseline large language models, particularly in the parsing of recursively nested clauses and the filtering of legal jargon. Resource consumption analysis indicates that although the agent framework results in a 40- to 60-fold increase in computational cost—consuming an average of 11,000 to 15,000 tokens per sentence—this overhead is justified in specific patent analysis scenarios due to its superior recall performance and risk mitigation effects. Furthermore, by classifying typical failure modes and proposing a hierarchical deployment strategy, this study provides a methodological framework and empirical foundation for deploying and applying the agent in specific technical domains.
- Open Access
- 11 July 2025
- Published by MDPI AG in Informatics
- Vol. 12 (3) , 69
A Patent-Based Technology Roadmap for AI-Powered Manipulators: An Evolutionary Analysis of the B25J Classification
Technology roadmapping is conducted by systematic mapping of technological evolution through patent analytics to inform innovation strategies. This study proposes an integrated framework combining hierarchical Latent Dirichlet Allocation (LDA) modeling with multiphase technology lifecycle theory, analyzing 113,449 Derwent patent abstracts (2008–2022) across three dimensions: technological novelty, functional applications, and competitive advantages. By segmenting innovation stages via logistic growth curve modeling and optimizing topic extraction through perplexity validation, we constructed dynamic technology roadmaps to decode latent evolutionary patterns in AI-powered programmable manipulators (B25J classification) within an innovation trajectory. Key findings revealed: (1) a progressive transition from electromechanical actuation to sensor-integrated architectures, evidenced by 58% compound annual growth in embedded sensing patents; (2) application expansion from industrial automation (72% early stage patents) to precision medical operations, with surgical robotics growing 34% annually since 2018; and (3) continuous advancements in adaptive control algorithms, showing 2.7× growth in reinforcement learning implementations. The methodology integrates quantitative topic modeling (via pyLDAvis visualization and cosine similarity analysis) with qualitative lifecycle theory, addressing the limitations of conventional technology analysis methods by reconciling semantic granularity with temporal dynamics. The results identify core innovation trajectories—precision control, intelligent detection, and medical robotics—while highlighting emerging opportunities in autonomous navigation and human–robot collaboration. This framework provides empirically grounded strategic intelligence for R&D prioritization, cross-industry investment, and policy formulation in Industry 4.0.
- 11 July 2025
- Published by MDPI AG in Informatics
- Vol. 12 (3) , 69
Open AccessA Patent-Based Technology Roadmap for AI-Powered Manipulators: An Evolutionary Analysis of the B25J Classification
Technology roadmapping is conducted by systematic mapping of technological evolution through patent analytics to inform innovation strategies. This study proposes an integrated framework combining hierarchical Latent Dirichlet Allocation (LDA) modeling with multiphase technology lifecycle theory, analyzing 113,449 Derwent patent abstracts (2008–2022) across three dimensions: technological novelty, functional applications, and competitive advantages. By segmenting innovation stages via logistic growth curve modeling and optimizing topic extraction through perplexity validation, we constructed dynamic technology roadmaps to decode latent evolutionary patterns in AI-powered programmable manipulators (B25J classification) within an innovation trajectory. Key findings revealed: (1) a progressive transition from electromechanical actuation to sensor-integrated architectures, evidenced by 58% compound annual growth in embedded sensing patents; (2) application expansion from industrial automation (72% early stage patents) to precision medical operations, with surgical robotics growing 34% annually since 2018; and (3) continuous advancements in adaptive control algorithms, showing 2.7× growth in reinforcement learning implementations. The methodology integrates quantitative topic modeling (via pyLDAvis visualization and cosine similarity analysis) with qualitative lifecycle theory, addressing the limitations of conventional technology analysis methods by reconciling semantic granularity with temporal dynamics. The results identify core innovation trajectories—precision control, intelligent detection, and medical robotics—while highlighting emerging opportunities in autonomous navigation and human–robot collaboration. This framework provides empirically grounded strategic intelligence for R&D prioritization, cross-industry investment, and policy formulation in Industry 4.0.
- 26 September 2024
- Published by Taylor & Francis in Information, Communication & Society
- p. 1-23
The thousand faces of images in AI news: psychological distance, dialectical relationships and sensationalism
- 26 September 2024
- Published by Taylor & Francis in Information, Communication & Society
- p. 1-23
The thousand faces of images in AI news: psychological distance, dialectical relationships and sensationalism
- 25 July 2024
- Published by SAGE Publications in Journal of Information Science
- Vol. 52 (4) , 1027-1040
All roads lead to Rome: Understanding the diffusion trajectories of innovation twins
This study conducts a comparison of the references and citations of two discoveries that were made simultaneously yet independently. The two discoveries, specifically ‘Inference of Population Structure Using Multilocus Genotype Data (IPSUMGD)’ and ‘Latent Dirichlet Allocation (LDA)’, are both significant academic publications in their respective fields. Although they share similar underlying concepts, they originate from different disciplines. Our objective is to analyse similarities and differences in the knowledge foundation and diffusion trajectories of these simultaneous discoveries, IPSUMGD and LDA, to further determine if a general pattern of successful innovation diffusion exists. The results indicate that the considerable similarity in the core ideas of IPSUMGD and LDA may be attributed to a strong disciplinary connection in their knowledge foundation, leading to overlapping diffusion processes. However, the divergence in thematic volatility and discipline distribution implies that IPSUMGD and LDA occupy distinct and independent diffusion spaces, which is crucial for their success. The citation cascade networks highlight the unique diffusion patterns of IPSUMGD and LDA, with IPSUMGD originating from the emergence of multiple high-impact nodes and LDA evolving through iterative innovation. The main path analysis reveals that both articles feature several key nodes in their diffusion processes, and the original authors have made substantial contributions to their long-term citation trajectories.
- 25 July 2024
- Published by SAGE Publications in Journal of Information Science
- Vol. 52 (4) , 1027-1040
All roads lead to Rome: Understanding the diffusion trajectories of innovation twins
This study conducts a comparison of the references and citations of two discoveries that were made simultaneously yet independently. The two discoveries, specifically ‘Inference of Population Structure Using Multilocus Genotype Data (IPSUMGD)’ and ‘Latent Dirichlet Allocation (LDA)’, are both significant academic publications in their respective fields. Although they share similar underlying concepts, they originate from different disciplines. Our objective is to analyse similarities and differences in the knowledge foundation and diffusion trajectories of these simultaneous discoveries, IPSUMGD and LDA, to further determine if a general pattern of successful innovation diffusion exists. The results indicate that the considerable similarity in the core ideas of IPSUMGD and LDA may be attributed to a strong disciplinary connection in their knowledge foundation, leading to overlapping diffusion processes. However, the divergence in thematic volatility and discipline distribution implies that IPSUMGD and LDA occupy distinct and independent diffusion spaces, which is crucial for their success. The citation cascade networks highlight the unique diffusion patterns of IPSUMGD and LDA, with IPSUMGD originating from the emergence of multiple high-impact nodes and LDA evolving through iterative innovation. The main path analysis reveals that both articles feature several key nodes in their diffusion processes, and the original authors have made substantial contributions to their long-term citation trajectories.
- 15 November 2023
- Published by Springer Nature
- p. 30-47
The Diffusion of Vaccine Hesitation: Media Visibility Versus Scientific Authority
- 15 November 2023
- Published by Springer Nature
- p. 30-47
The Diffusion of Vaccine Hesitation: Media Visibility Versus Scientific Authority
- 27 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. PP (10414347) , 1-6
Interpretable Stochastic Block Influence Model: Measuring Social Influence Among Homophilous Communities
Decision-making on networks can be explained by both homophily and social influences. While homophily drives the formation of communities with similar characteristics, social influences occur both within and between communities. Social influences can be reasoned through role theory, which indicates that the influences among individuals depending on their roles and the behavior of interest. To operationalize these social science theories, we empirically identify the homophilous communities and use the community structures to capture such “roles”, affecting particular decision-making processes. We propose a generative model named the Stochastic Block influences Model and jointly analyzed both network formation and behavioral influences within and between different empirically-identified communities. To evaluate the performance and demonstrate the interpretability of our method, we study the adoption decisions for a microfinance product in Indian villages. We show that although individuals tend to form links within communities, there are strongly positive and negative social influences between communities, supporting the weak ties theory. Moreover, communities with shared characteristics are associated with positive influences. In contrast, communities that do not overlap are associated with negative influences. Our framework facilitates the quantification of the influences underlying decision communities and is thus a helpful tool for driving information diffusion, viral marketing, and technology adoption.
- 27 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. PP (10414347) , 1-6
Interpretable Stochastic Block Influence Model: Measuring Social Influence Among Homophilous Communities
Decision-making on networks can be explained by both homophily and social influences. While homophily drives the formation of communities with similar characteristics, social influences occur both within and between communities. Social influences can be reasoned through role theory, which indicates that the influences among individuals depending on their roles and the behavior of interest. To operationalize these social science theories, we empirically identify the homophilous communities and use the community structures to capture such “roles”, affecting particular decision-making processes. We propose a generative model named the Stochastic Block influences Model and jointly analyzed both network formation and behavioral influences within and between different empirically-identified communities. To evaluate the performance and demonstrate the interpretability of our method, we study the adoption decisions for a microfinance product in Indian villages. We show that although individuals tend to form links within communities, there are strongly positive and negative social influences between communities, supporting the weak ties theory. Moreover, communities with shared characteristics are associated with positive influences. In contrast, communities that do not overlap are associated with negative influences. Our framework facilitates the quantification of the influences underlying decision communities and is thus a helpful tool for driving information diffusion, viral marketing, and technology adoption.
- Open Access
- 14 October 2022
- Published by MDPI AG in International Journal of Environmental Research and Public Health
- Vol. 19 (20) , 13287
COVID-19 Vaccines on TikTok: A Big-Data Analysis of Entangled Discourses
Focusing on social media affordances and China’s social/political context, the present study analyzed the digital communication practices about COVID-19 vaccines on a popular social media platform—TikTok—which is called DouYin in China. Overall, this study identified five major forces partaking in constructing the discourses, with government agencies and state media being the dominant contributors. Furthermore, video posters demonstrated different patterns of utilizing social media affordances (e.g., hashtags) in disseminating their messages. The top hashtags adopted by state media were more representative of international relations and Taiwan; those by government agencies were of updates on pandemic outbreaks; those by individual accounts were of mainstream values and health education; those by commercial media were of celebrities and health education; those by enterprise accounts were of TikTok built-in marketing hashtags. The posted videos elicited both cognitive and affective feedback from online viewers. Implications of the findings were discussed in the context of health communication and global recovery against the backdrop of the COVID-19 pandemic and Chinese culture.
- 14 October 2022
- Published by MDPI AG in International Journal of Environmental Research and Public Health
- Vol. 19 (20) , 13287
Open AccessCOVID-19 Vaccines on TikTok: A Big-Data Analysis of Entangled Discourses
Focusing on social media affordances and China’s social/political context, the present study analyzed the digital communication practices about COVID-19 vaccines on a popular social media platform—TikTok—which is called DouYin in China. Overall, this study identified five major forces partaking in constructing the discourses, with government agencies and state media being the dominant contributors. Furthermore, video posters demonstrated different patterns of utilizing social media affordances (e.g., hashtags) in disseminating their messages. The top hashtags adopted by state media were more representative of international relations and Taiwan; those by government agencies were of updates on pandemic outbreaks; those by individual accounts were of mainstream values and health education; those by commercial media were of celebrities and health education; those by enterprise accounts were of TikTok built-in marketing hashtags. The posted videos elicited both cognitive and affective feedback from online viewers. Implications of the findings were discussed in the context of health communication and global recovery against the backdrop of the COVID-19 pandemic and Chinese culture.
- 23 February 2022
- Published by Springer Nature
- p. 61-68
Elapsed Collective Memory: Looking for the Forgotten Classic Works in Library and Information Science
- 23 February 2022
- Published by Springer Nature
- p. 61-68
Elapsed Collective Memory: Looking for the Forgotten Classic Works in Library and Information Science
- 17 January 2022
- Published by Wiley in European Eating Disorders Review
- Vol. 30 (2) , 96-109
The landscape of eating disorders research: A 40‐year bibliometric analysis
Objective: Employing bibliometric methods, the present study aimed to map out the general landscape of existing research on eating disorders (EDs) over the past decades.Method: Using the Web of Science database, we retrieved 41,917 research articles related to EDs published from 1981 to 2020. After removing those without an abstract, a total of 37,446 articles were retained. The study outlined the distribution of scholarship by time, languages, regions, and countries, and identified major research lines by applying latent topic modelling.Results: Results revealed a general increasing trend in the number of publications on EDs research, and researchers from Western countries dominated the production of related scholarship. The distribution of published scholarship varied significantly by languages, regions, and countries. Seven main research topics emerged from past research (i.e., animal studies of food intake, risk factors and at‐risk groups for eating disorders, body image in eating disorders, studies of cognition and brain in eating disorders, symptomatology and comorbidity of eating disorders, body weight and nutrition status in eating disorders, and treatment of eating disorders), with different topics showing unique research trends across the years.Conclusions: This bibliometric analysis presents the most complete up‐to‐date overview on published research on EDs. While there is an increasing trend for EDs research, the available research evidence is generally from Western countries; thus, it is suggested that cooperation on EDs research should be strengthened between Western countries and other countries in the future.
- 17 January 2022
- Published by Wiley in European Eating Disorders Review
- Vol. 30 (2) , 96-109
The landscape of eating disorders research: A 40‐year bibliometric analysis
Objective: Employing bibliometric methods, the present study aimed to map out the general landscape of existing research on eating disorders (EDs) over the past decades.Method: Using the Web of Science database, we retrieved 41,917 research articles related to EDs published from 1981 to 2020. After removing those without an abstract, a total of 37,446 articles were retained. The study outlined the distribution of scholarship by time, languages, regions, and countries, and identified major research lines by applying latent topic modelling.Results: Results revealed a general increasing trend in the number of publications on EDs research, and researchers from Western countries dominated the production of related scholarship. The distribution of published scholarship varied significantly by languages, regions, and countries. Seven main research topics emerged from past research (i.e., animal studies of food intake, risk factors and at‐risk groups for eating disorders, body image in eating disorders, studies of cognition and brain in eating disorders, symptomatology and comorbidity of eating disorders, body weight and nutrition status in eating disorders, and treatment of eating disorders), with different topics showing unique research trends across the years.Conclusions: This bibliometric analysis presents the most complete up‐to‐date overview on published research on EDs. While there is an increasing trend for EDs research, the available research evidence is generally from Western countries; thus, it is suggested that cooperation on EDs research should be strengthened between Western countries and other countries in the future.
- 29 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 7 (1) , 69-80
Misinformation During the COVID-19 Outbreak in China: Cultural, Social and Political Entanglements
Not only did COVID-19 give rise to a global pandemic, but also it resulted in an infodemic. The consequences of this infodemic can erode public trust and outlive the pandemic. The evolving and fragmented media landscape, particularly the extensive use of social media, is a crucial driver of the spread of misinformation. We collected four million Weibo posts on the Chinese social media from 12/2019 to 4/2020, examining misinformation identified by a fact-checking platform. Our results show that the evolution of misinformation follows an issue-attention cycle, pertaining to topics such as city lockdown, cures, and school-reopening. Sensational-reassuring misinformation characterizes the whole issue-attention cycle, with misinformation on cures and prevention flooding social media. We also study the evolution of sentiment and observe that positive sentiment dominated during COVID-19, which may be due to the unique characteristic of "positive energy'' on Chinese social media. Lastly, we study the media landscape during Covid via a case study on a controversial unproven cure, which testifies the importance of scientific communication in a plague. Our findings shed light on misinformation's distinct characteristics and its cultural, social, and political implications during the COVID-19. The study offers insights into combating misinformation in China and across the world.
- 29 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 7 (1) , 69-80
Misinformation During the COVID-19 Outbreak in China: Cultural, Social and Political Entanglements
Not only did COVID-19 give rise to a global pandemic, but also it resulted in an infodemic. The consequences of this infodemic can erode public trust and outlive the pandemic. The evolving and fragmented media landscape, particularly the extensive use of social media, is a crucial driver of the spread of misinformation. We collected four million Weibo posts on the Chinese social media from 12/2019 to 4/2020, examining misinformation identified by a fact-checking platform. Our results show that the evolution of misinformation follows an issue-attention cycle, pertaining to topics such as city lockdown, cures, and school-reopening. Sensational-reassuring misinformation characterizes the whole issue-attention cycle, with misinformation on cures and prevention flooding social media. We also study the evolution of sentiment and observe that positive sentiment dominated during COVID-19, which may be due to the unique characteristic of "positive energy'' on Chinese social media. Lastly, we study the media landscape during Covid via a case study on a controversial unproven cure, which testifies the importance of scientific communication in a plague. Our findings shed light on misinformation's distinct characteristics and its cultural, social, and political implications during the COVID-19. The study offers insights into combating misinformation in China and across the world.